Looking for breakthrough ideas for innovation challenges? Try Patsnap Eureka!

Power transmission line defect detection method based on hierarchical region feature fusion learning

A technology of regional characteristics and transmission lines, which is applied in the field of image analysis, can solve problems such as complex and diverse defects and target background differences, and achieve the effects of strengthening overall perception capabilities, improving efficiency, and saving time for adjusting parameters

Active Publication Date: 2019-10-15
NORTH CHINA ELECTRIC POWER UNIV (BAODING) +4
View PDF6 Cites 40 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0008] The purpose of the present invention is to overcome the deficiencies of the prior art, provide a transmission line defect detection method based on hierarchical regional feature fusion learning, and solve the complexity and diversity of defects and the huge difference in target background in the existing transmission line defect detection

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Power transmission line defect detection method based on hierarchical region feature fusion learning
  • Power transmission line defect detection method based on hierarchical region feature fusion learning
  • Power transmission line defect detection method based on hierarchical region feature fusion learning

Examples

Experimental program
Comparison scheme
Effect test

Embodiment 1

[0075] Transmission line defect detection method based on hierarchical regional feature fusion learning, please refer to the attached figure 1 As shown, including: constructing and calling the Faster R-CNN model; regressing the target features extracted by the backbone network through the RPN network to obtain the target area; performing RoI pooling operations on the input image to generate local deep and shallow regional features, and generating them through the depth selection network The weight required for feature fusion fuses the deep feature area and the shallow feature area; and generates the final prediction result through the classification network and the regression network.

[0076] The specific steps are as follows:

[0077] S1. ImageNet-based network pre-training:

[0078] Use the vgg16 network as the deep feature extraction network, pre-train on the ImageNet data set, and use the weight value obtained from the pre-training as the initial parameter value of the m...

Embodiment 2

[0124] The inspection line image is used as the input of the depth detection model, and the pre-trained vgg16 network extracts features, and the depth model generates anchors of different sizes (the size of the anchors is set to (4×4), (8×8), (16 ×16), (32×32), the aspect ratios are 0.5, 1, 2), the region of interest is calculated through the RPN network prediction offset, and the deep image features generated by the vgg16 network are matched with the image shallow features of the feature channel. The layer features are respectively sent to the RoI Pooling layer to obtain deep feature ROIs and shallow feature ROIs. The weight of feature fusion is obtained by the depth selection network, and the fused feature map is generated by the method of weighted sum, and the feature map is visualized to obtain the region-level fusion feature map, such as figure 2 shown.

Embodiment 3

[0126] In this embodiment, on the basis of Embodiment 1 and Embodiment 2, the fusion features obtained in Embodiment 2 are used as the input of the classification network and the regression network respectively, and the category score and bounding box offset are predicted, and the cross-entropy loss function and Smooth L 1 The loss function calculates the loss value and learns through backpropagation. At this time, the batchsize is set to 128, the learning rate decay is set to 0.001, the number of iterations is 40,000, and the candidate frame is selected using the non-maximum suppression method. After the training is completed, input the inspection image to be detected into the depth model, and get image 3 Detected images labeled with predicted classes and predicted bounding boxes in .

[0127] Based on the Faster R-CNN model, the present invention returns the target features extracted by the backbone network through the RPN network to obtain the target area, and performs Ro...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

The invention discloses a power transmission line defect detection method based on hierarchical region feature fusion learning. The power transmission line defect detection method comprises the stepsof constructing and calling a Faster R-CNN model; performing RPN network regression on the target features extracted by the backbone network to obtain a target area; carrying out roI pooling operationon an input image to generate local hierarchical region features, and generating weights required by feature fusion through deep selection network learning to fuse a deep feature region and a shallowfeature region; and generating a final prediction result through the classification network and the regression network. According to the invention, a self-learning regional feature fusion weight is generated by using a deep selection network; saving time to adjust parameters, fusion features obtained by model learning can better adapt to defect detection tasks under different complex conditions;the depth model performs prediction by using the regional features, enhances the learning ability of the model for extracting the local features of the target, and reduces the false detection problemof the model in the actual environment due to the complex background of the defect image of the power transmission line and the inter-class difference.

Description

technical field [0001] The invention relates to the technical field of image analysis, in particular to a transmission line defect detection method based on hierarchical region feature fusion learning. Background technique [0002] As the carrier of long-distance electric energy transmission, the transmission line is in the harsh field environment for a long time, and is easily damaged by wind, rain, snow, animals and other factors. In severe cases, it will cause large-scale power outages and incalculable economic losses. Therefore, fine inspection and maintenance of transmission lines has become an important task in the current power system. At present, aircraft inspection has become a routine inspection method, which has high efficiency and low economic cost. However, the main pain point of inspection is the contradiction between the rapid and substantial increase in the demand for aerial image defect detection and the relatively low accuracy and efficiency of manual dete...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
Patent Type & Authority Applications(China)
IPC IPC(8): G06T7/00G06K9/46G06K9/32G06K9/62G06N3/04G06N3/08G01N21/88G01R31/08
CPCG06T7/0004G06N3/084G01N21/8851G01R31/088G01R31/085G06T2207/10004G06T2207/20104G06T2207/20081G06T2207/30108G01N2021/8887G01N2021/8883G06V10/25G06V10/462G06N3/045G06F18/214G06F18/241Y04S10/50
Inventor 赵振兵李延旭戚银城赵文清
Owner NORTH CHINA ELECTRIC POWER UNIV (BAODING)
Who we serve
  • R&D Engineer
  • R&D Manager
  • IP Professional
Why Patsnap Eureka
  • Industry Leading Data Capabilities
  • Powerful AI technology
  • Patent DNA Extraction
Social media
Patsnap Eureka Blog
Learn More
PatSnap group products

Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic, Popular Technical Reports.

© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com